HKBU  |  SCI  |  BUniPort  |  Library  |  Alumni  |  Job Vacancies  |  Intranet  |  Sitemap        
Undergraduate Admissions
Taught Postgraduate Admissions
Research Postgraduate Admissions
News & Achievements
Research Highlights
Contact & Direction

Department of Computer Science Colloquium
2007 Series

Self-Organising Map Techniques for Graph-Data Applications

Dr. Markus Hagenbuchner
School of Computer Science and Software Engineering
University of Wollongong

Date: May 16, 2007 (Wednesday)
Time: 2:30 - 3:30 pm
Venue: RRS905, Sir Run Run Shaw Building, Ho Sin Hang Campus

This talk presents neural network techniques based on Kohonen's self-organising map method which can be trained in an unsupervised fashion, and which are applicable to the processing of graph structured inputs. It is shown how such techniques can be applied to a variety of learning problems such as subgraph matching, clustering and labeling of structured data, pattern detection, and more. Neural networks are popularly applied machine learning techniques. There are a number of popular network architectures, e.g. multilayer perceptrons, self organising maps, support vector machines. However, most of these techniques have been applied to problems in form of fixed sized inputs. In other words, the inputs to these neural network architectures are expressed in the form of vectors. When dealing with application domains were inputs are not normally expressed in the form of vectors, they are made to conform to the fixed dimension vectorial format. For example, it is known that an image may be more appropriately expressed in the form of a graph, for instance, chemical molecules can be expressed as a graph, with nodes representing the atoms and links represent the atomic bindings. Nodes and links can be described by attributes (features such as the atomic weight, strength and type of an atomic bond). Such inputs can be made to conform to a vectorial format if we "flatten" the structure and instead represent the information in each node in the form of a vector, and obtain the aggregate vector by concatenating the vectors together. Such techniques have been prevalent in the application of neural network architectures to these problems. Another way to process the data is to preserve the graph structured data format, and modify the neural network techniques to process graph structured data. In this talk, we will address ways to modify a classic neural network technique, Self-Organising Maps, so that it can accept graph structured inputs.

Markus Hagenbuchner holds a PhD (Computer Science, University of Wollongong, Australia). He is currently a lecturer in the School of Computer Science and Software Engineering at the University of Wollongong. He joint the machine learning research area in 1992, and started to focus his research activities on Neural Networks for the graph structured domain in 1998. He is the team leader of the machine learning group at the University of Wollongong.

********* ALL INTERESTED ARE WELCOME ***********
(For enquiry, please contact Computer Science Department at 3411 2385)
Copyright © 2021. All rights reserved.Privacy Policy
Department of Computer Science, Hong Kong Baptist University
Hong Kong Baptist University